We conclude this series with a look at control schemes that can increase process efficiency for distillation control, compressor control, and neutralizer control. These control schemes can inherently reduce process variability and controller tuning difficulty and for compressors increase onstream time. Process control improvements can be ready for operator training by the use of a virtual plant to test the configuration for a full spectrum of process conditions including abnormal situations in a matter of days.
For insight into what are the most effective column control schemes we look at Terry Tolliver’s ISA Automation Week 2011 Tutorial ISA-AW-2011-Fundamentals-of-Distillation-Column-Control.pdf. Material balance control (slide 16) and separation control (slide 17) are the two general types of control. In separation control, the column temperature manipulates the energy input (e.g. steam) and one of the exit streams (distillate or bottoms flow) is on flow control rather than level control. Material balance control manipulates directly by temperature control or indirectly by level control each of the output flows (distillate and bottoms flow). Material balance control provides the broadest range of setpoint selection and disturbance rejection as seen on slide 15. By changing the distillate to feed ratio (D/F) in material balance control, you can affect nearly the full range of light key component concentration in the distillate (mole fraction y) and in the bottoms (mole fraction x). Furthermore, you can deal with almost the entire possible range of light key component in the feed composition (mole fraction z). On this same slide 15 we see that changing the separation at any given feed concentration, provides a narrow band of possible changes in the concentration in the distillate and bottoms. Thus, separation control is generally not recommended except for high purity columns where you are operating with very little light component key in the bottoms or very little heavy component key in the distillate. In these cases, in the tail end of the plot of component concentration versus the ratio of manipulated flow to feed flow (D/F for light component and B/F for heavy component) the slope flattens showing a loss of sensitivity.
There are four major types of material balance control. Type 1 (slide 21) directly adjusts the material balance by the temperature controller setting distillate flow. Type 1 has the least interaction with the energy balance and provides some degree of internal reflux control if the distillate receiver level controller is tightly tuned. For example, if a rain storm cools off the column and overheads walls causing the reflux to be colder and more condensing in the column, the overhead head vapor flow is reduced. Tight level controller will then reduce the reflux flow to help compensate for the colder reflux. This scheme is the first choice if the distillate flow is one of the smaller flows in the column and hence the reflux flow is high.
If the reflux flow is small compared to the distillate flow (R/D < 1), then Type 2 (slide 22) indirect material balance control is more effective where the temperature controller manipulates the reflux flow and the distillate level controller manipulates reflux flow. The level control does not need to be tight allowing some room to smooth distillate disturbances to downstream equipment.
If the vapor to feed ratio is low (V/F < 2), then Type 3 (slide 23) indirect material balance control where the temperature manipulates the steam flow is a viable solution. This scheme has the fastest response because vapors proceed up the column rapidly but there is a greater interaction with the energy balance. The bottoms level control does not need to be particularly tightly controlled. However, the inventory in the sump may be relatively low requiring aggressive level control to keep the reboiler pump net positive suction head (NPSH) dropping too low and the reboiler being starved from low sump level or obstruction of vapor flow from high sump level.
If the bottoms to feed flow ratio is low (B/F < 1), then Type 4 (slide 24) direct material control balance control where the temperature manipulates the bottoms flow may be necessary. This scheme is problematic in that tight level control is desirable as with Type 1 but here the use of steam can create inverse response that necessitates detuning of the sump level controller. A flow feedforward added to the sump level controller output to maintain a V/F ratio can help considerably.
A simplistic guide as to which Type control system to use would be to select the one where the temperature controller manipulate a liquid flow that is less than the feed flow and a vapor flow that is less than twice the feed flow. The distillate and sump level controllers are then manipulating a flow large enough to deal with feed disturbances.
In general, one of the biggest opportunities for distillation control is ratio control of column flows to feed flow (D/F, R/F, V/F, and B/F). Columns are commonly started up on ratio control with temperature controller in manual until the column has reached operating conditions. These ratios should be displayed and adjustable on the operator interface. Dynamic compensation may be needed to improve the timing of the flow feedforward signals. In general a feedforward summer is used rather than a feedforward multiplier to reduce feedforward correction scaling issues and to compensate for offset errors in the measurement and base loading of the column.
Greg Shinskey also heavily advocates the use of material balance control instead of separation control. In his book Distillation Control for Productivity and Energy Conservation Greg offers a detailed relative gain analysis for the various control schemes as well as a wealth of very practical information. Unfortunately, nearly all of Shinskey’s books are out of print. A few used copies are available on "Shinskey's Page" on Amazon.com. Peter Martin, Automation Week 2012 program chair, graciously sent me a book signed by Shinskey after seeing me lament not having a copy in my recorded interview of Shinskey for ISA Automation week 2011. I now have all of Shinskey’s books and don’t have to deal with the escalating prices of used copies. For the life of me I don’t understand how publishers can let them go out of print. I guess they don’t know these books detail knowledge that can never be totally and exactly replaced.
For compressor control, the idea of the game is to prevent surge since each surge cycle causes a loss in compressor efficiency from seal damage. The damage may be ever so slight but still exists and eventually shows up in loss in efficiency and maintenance. In the case of some axial compressors, the damage from just a single surge cycle is detectable and successive surge cycles can cause rotor damage. The interruption of flow from surge or just the opening of the surge valve often ends up in the shutdown of downstream users causing loss in production rate and more difficult and potentially hazardous operation during shutdown and startup.
Feedforward of downstream user flows is tricky due to the nonlinear installed characteristic of the surge valves. A signal characterizer on the PID may be the solution and would act to provide a faster besides more linear manipulation initial opening of the surge valve for an equal percentage characteristic. For checking the operation of the surge valve, the characterized signal besides the raw PID output needs to be displayed.
The prevention of surge is best accomplished by the use of a precise and fast automation system and a surge setpoint that is an accurate and compensated for compressor age and i some cases molecular weight if the suction gas composition is not constant. Incredibly fast surge valves (pre-stroke deadtime < 0.1 sec and stroking time < 1 sec) and fast transmitters (damping < 0.2 sec) are needed. If air actuated surge valves are used, a PID execution time of 0.1 sec is sufficient (< 10% of reset time as noted in 2-22-2012 Control Talk Blog). If hydraulic and stepper motor or solenoid valve arrays are used, faster PID execution rates are needed to take advantage of the faster surge valves.
Surge is the fastest known process disturbance with the initial precipitous drop in flow occurring in milliseconds. The momentum balance is too fast for most real time dynamic simulations. Virtual plants require the slowing down of the process and the control system to simulate and analyze the consequences. The book Centrifugal and Axial Compressor Control provides the equations, understanding, and program listing for the simulation of surge using a momentum balance.
Once a compressor gets into cycling, feedback control may not be able to get the compressor out of surge. A directional velocity limit in the analog output (AO) block to provide a fast opening but slow closing with external reset feedback to keep the controller reset in sync can help considerably. At any rate, an open loop backup or kicker is employed to rapidly move the surge valve open if the onset of surge is detected. The operating point when surge is detected can be used to update the surge curve by a bias that is a fraction of the difference between the actual and predicted surge point. The use of an intelligent detection by the simple use of a deadtime block to create a delta flow over the loop deadtime (another example of the general technique mentioned at the end of Part 3) can prevent surge and reduce unnecessary opening of the surge valve.
The tuning of the controller must not cause an unnecessary opening of the surge valve. The relative contribution of integral to gain action needs to be adjusted while the PID output is at its output limit to enable the surge valve to open just in time with minimal undershoot. Any biasing (shifting) of the surge curve to allow for undershoot should be considered in the analysis. The integral mode tries to prevent the surge valve from opening until the flow has dropped below the surge setpoint. The proportional mode causes the valve to open sooner. Increasing gain action and decreasing reset action will cause the surge valve to open sooner. The reset time can be adapted to enable the valve to start to open at a desired flow relative to the surge curve by looking at rate at which the compressor flow is dropping using the simple deadtime block configuration described at the end of Part 3 to project a point where the surge valve needs to open relative to the surge curve and the contribution of the proportional mode needs to exceed the contribution from the integral mode. Included must be the effect of the PID algorithms at the output limit. For example, in some algorithms reset action is a factor of ten larger (reset time a factor of ten smaller) between the output and anti-reset windup limit.
This same open loop backup or kicker technique is also used to prevent a RCRA pH violation. If delta pH over the loop deadtime shows the pH one deadtime in the future is getting too close (again using the technique described at the end of Part 3) to the RCRA limit, the reagent valve is incrementally opened. As with the surge valve, unnecessary opening of the control valve should be minimized for process efficiency. As soon as the project condition clears the incremental opening of the reagent valve stops and is turned back over to pH control. Here again an AO directional velocity limit with external reset feedback can provide a fast getaway from trouble and a slow approach to the optimum setpoint. A recordable RCRA violation is undesirable from many aspects the most severe triggering an environmental review. The November 2011 control magazine article "Virtual Control of Real pH" showed how the use of virtual plant lead to simple improvements in the kicker design that saved several hundred thousand of dollars in reagent each year.
Finally, for pH control the use of static mixers and signal characterization can reduce process variability as described in slides 61 and 68 of the ISA Automation Week 2011 tutorial pH-Measurement-and-Control-Opportunities.pdf and in the Chemical Processing article "Virtual Plant Provides Real Insights"